Massive data are generated by social media users including posts, tweets, images, and videos. Getting valuable information from this big data is a significant, challenging, and interesting issue in the text mining domain. Twitter data are analyzed with text mining techniques to discover society agenda, trends, user behaviors, and feelings. Text analysis method to determine sentiments from tweets is proposed in the present research. Apache Flume is used to collect data stream from Twitter and store into Apache Hadoop. Natural language processing techniques are carried out to put the data into meaningful context followed by a classification model training with data mining methods. It carries out the classification label as people’s opinion, such as positive, negative, and neutral sentiments, using Twitters streaming data. 10 different automobile brands are selected and collected tweets with hashtags about these brands by using Apache Flume are used as case study. Collected data have been pre-processed using TF-IDF, Bi-gram, and SVD metrics and a classification tree model has been generated and the results are compared. The results that were experimented indicated that the classification tree based on SVD has the best accuracy. According to the different brands model, based on bigram is the most stable and performs with the best accuracy. The results from the experiments indicate that the model that uses Bi gram could be used to address data with complex behavior in the sentiment detection.